Along with advanced informatization of current social systems, the industrial application of pattern classification technologies, such as voice recognition and image recognition has been attracting attention. However, since electronic information measured in daily life contains uncertainties such as noise, such products are required that can demonstrate consistently high discrimination capability even under such unfavorable conditions.
Our laboratory is developing a neural network system (Japanese patent 2002-366927) that discriminates data stochastically by empirically extracting statistical characteristics from sample data that contains uncertainties. A neutral network (NN) is capable of discriminating data for which characteristics change in a time-series manner. With this system, learning time can be specified, which has been impossible with conventional NNs.
Since the NN we have developed is capable of classifying patterns of various data, a versatile pattern classification system can be developed by leveraging this characteristic.